net/virtio: replace full barrier with relaxed ones for Arm
[dpdk.git] / drivers / net / virtio / virtio_ethdev.c
index 6e79a26..92a3d4e 100644 (file)
@@ -14,7 +14,6 @@
 #include <rte_string_fns.h>
 #include <rte_memzone.h>
 #include <rte_malloc.h>
-#include <rte_atomic.h>
 #include <rte_branch_prediction.h>
 #include <rte_pci.h>
 #include <rte_bus_pci.h>
 #include <rte_common.h>
 #include <rte_errno.h>
 #include <rte_cpuflags.h>
+#include <rte_vect.h>
 
 #include <rte_memory.h>
 #include <rte_eal.h>
 #include <rte_dev.h>
 #include <rte_cycles.h>
+#include <rte_kvargs.h>
 
 #include "virtio_ethdev.h"
 #include "virtio_pci.h"
 #include "virtio_logs.h"
 #include "virtqueue.h"
 #include "virtio_rxtx.h"
+#include "virtio_user/virtio_user_dev.h"
 
 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
 static int  virtio_dev_configure(struct rte_eth_dev *dev);
 static int  virtio_dev_start(struct rte_eth_dev *dev);
-static void virtio_dev_stop(struct rte_eth_dev *dev);
-static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
-static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
-static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
-static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
-static void virtio_dev_info_get(struct rte_eth_dev *dev,
+static int  virtio_dev_stop(struct rte_eth_dev *dev);
+static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
+static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
+static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
+static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
+static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
+static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
+       int *vdpa,
+       uint32_t *speed,
+       int *vectorized);
+static int virtio_dev_info_get(struct rte_eth_dev *dev,
                                struct rte_eth_dev_info *dev_info);
 static int virtio_dev_link_update(struct rte_eth_dev *dev,
        int wait_to_complete);
@@ -60,18 +67,17 @@ static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
                                       struct rte_eth_xstat_name *xstats_names,
                                       unsigned limit);
-static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
+static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
                                uint16_t vlan_id, int on);
 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
-                               struct ether_addr *mac_addr,
+                               struct rte_ether_addr *mac_addr,
                                uint32_t index, uint32_t vmdq);
 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
-static void virtio_mac_addr_set(struct rte_eth_dev *dev,
-                               struct ether_addr *mac_addr);
+static int virtio_mac_addr_set(struct rte_eth_dev *dev,
+                               struct rte_ether_addr *mac_addr);
 
-static int virtio_intr_enable(struct rte_eth_dev *dev);
 static int virtio_intr_disable(struct rte_eth_dev *dev);
 
 static int virtio_dev_queue_stats_mapping_set(
@@ -80,9 +86,6 @@ static int virtio_dev_queue_stats_mapping_set(
        uint8_t stat_idx,
        uint8_t is_rx);
 
-int virtio_logtype_init;
-int virtio_logtype_driver;
-
 static void virtio_notify_peers(struct rte_eth_dev *dev);
 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
 
@@ -121,7 +124,6 @@ static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
        {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
        {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
-       {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
        {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
        {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
        {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
@@ -141,38 +143,112 @@ static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 
 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 
-static int
-virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
-               int *dlen, int pkt_num)
+static struct virtio_pmd_ctrl *
+virtio_send_command_packed(struct virtnet_ctl *cvq,
+                          struct virtio_pmd_ctrl *ctrl,
+                          int *dlen, int pkt_num)
 {
-       uint32_t head, i;
-       int k, sum = 0;
-       virtio_net_ctrl_ack status = ~0;
+       struct virtqueue *vq = cvq->vq;
+       int head;
+       struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
        struct virtio_pmd_ctrl *result;
-       struct virtqueue *vq;
+       uint16_t flags;
+       int sum = 0;
+       int nb_descs = 0;
+       int k;
 
-       ctrl->status = status;
+       /*
+        * Format is enforced in qemu code:
+        * One TX packet for header;
+        * At least one TX packet per argument;
+        * One RX packet for ACK.
+        */
+       head = vq->vq_avail_idx;
+       flags = vq->vq_packed.cached_flags;
+       desc[head].addr = cvq->virtio_net_hdr_mem;
+       desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
+       vq->vq_free_cnt--;
+       nb_descs++;
+       if (++vq->vq_avail_idx >= vq->vq_nentries) {
+               vq->vq_avail_idx -= vq->vq_nentries;
+               vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
+       }
 
-       if (!cvq || !cvq->vq) {
-               PMD_INIT_LOG(ERR, "Control queue is not supported.");
-               return -1;
+       for (k = 0; k < pkt_num; k++) {
+               desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
+                       + sizeof(struct virtio_net_ctrl_hdr)
+                       + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
+               desc[vq->vq_avail_idx].len = dlen[k];
+               desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
+                       vq->vq_packed.cached_flags;
+               sum += dlen[k];
+               vq->vq_free_cnt--;
+               nb_descs++;
+               if (++vq->vq_avail_idx >= vq->vq_nentries) {
+                       vq->vq_avail_idx -= vq->vq_nentries;
+                       vq->vq_packed.cached_flags ^=
+                               VRING_PACKED_DESC_F_AVAIL_USED;
+               }
        }
 
-       rte_spinlock_lock(&cvq->lock);
-       vq = cvq->vq;
-       head = vq->vq_desc_head_idx;
+       desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
+               + sizeof(struct virtio_net_ctrl_hdr);
+       desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
+       desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
+               vq->vq_packed.cached_flags;
+       vq->vq_free_cnt--;
+       nb_descs++;
+       if (++vq->vq_avail_idx >= vq->vq_nentries) {
+               vq->vq_avail_idx -= vq->vq_nentries;
+               vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
+       }
 
-       PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
-               "vq->hw->cvq = %p vq = %p",
-               vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
+       virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
+                       vq->hw->weak_barriers);
 
-       if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
-               rte_spinlock_unlock(&cvq->lock);
-               return -1;
-       }
+       virtio_wmb(vq->hw->weak_barriers);
+       virtqueue_notify(vq);
 
-       memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
-               sizeof(struct virtio_pmd_ctrl));
+       /* wait for used desc in virtqueue
+        * desc_is_used has a load-acquire or rte_io_rmb inside
+        */
+       while (!desc_is_used(&desc[head], vq))
+               usleep(100);
+
+       /* now get used descriptors */
+       vq->vq_free_cnt += nb_descs;
+       vq->vq_used_cons_idx += nb_descs;
+       if (vq->vq_used_cons_idx >= vq->vq_nentries) {
+               vq->vq_used_cons_idx -= vq->vq_nentries;
+               vq->vq_packed.used_wrap_counter ^= 1;
+       }
+
+       PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
+                       "vq->vq_avail_idx=%d\n"
+                       "vq->vq_used_cons_idx=%d\n"
+                       "vq->vq_packed.cached_flags=0x%x\n"
+                       "vq->vq_packed.used_wrap_counter=%d\n",
+                       vq->vq_free_cnt,
+                       vq->vq_avail_idx,
+                       vq->vq_used_cons_idx,
+                       vq->vq_packed.cached_flags,
+                       vq->vq_packed.used_wrap_counter);
+
+       result = cvq->virtio_net_hdr_mz->addr;
+       return result;
+}
+
+static struct virtio_pmd_ctrl *
+virtio_send_command_split(struct virtnet_ctl *cvq,
+                         struct virtio_pmd_ctrl *ctrl,
+                         int *dlen, int pkt_num)
+{
+       struct virtio_pmd_ctrl *result;
+       struct virtqueue *vq = cvq->vq;
+       uint32_t head, i;
+       int k, sum = 0;
+
+       head = vq->vq_desc_head_idx;
 
        /*
         * Format is enforced in qemu code:
@@ -180,30 +256,30 @@ virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
         * At least one TX packet per argument;
         * One RX packet for ACK.
         */
-       vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
-       vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
-       vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
+       vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
+       vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
+       vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
        vq->vq_free_cnt--;
-       i = vq->vq_ring.desc[head].next;
+       i = vq->vq_split.ring.desc[head].next;
 
        for (k = 0; k < pkt_num; k++) {
-               vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
-               vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
+               vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
+               vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
                        + sizeof(struct virtio_net_ctrl_hdr)
                        + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
-               vq->vq_ring.desc[i].len = dlen[k];
+               vq->vq_split.ring.desc[i].len = dlen[k];
                sum += dlen[k];
                vq->vq_free_cnt--;
-               i = vq->vq_ring.desc[i].next;
+               i = vq->vq_split.ring.desc[i].next;
        }
 
-       vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
-       vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
+       vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
+       vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
                        + sizeof(struct virtio_net_ctrl_hdr);
-       vq->vq_ring.desc[i].len = sizeof(ctrl->status);
+       vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
        vq->vq_free_cnt--;
 
-       vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
+       vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
 
        vq_update_avail_ring(vq, head);
        vq_update_avail_idx(vq);
@@ -212,28 +288,26 @@ virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 
        virtqueue_notify(vq);
 
-       rte_rmb();
-       while (VIRTQUEUE_NUSED(vq) == 0) {
-               rte_rmb();
+       while (virtqueue_nused(vq) == 0)
                usleep(100);
-       }
 
-       while (VIRTQUEUE_NUSED(vq)) {
+       while (virtqueue_nused(vq)) {
                uint32_t idx, desc_idx, used_idx;
                struct vring_used_elem *uep;
 
                used_idx = (uint32_t)(vq->vq_used_cons_idx
                                & (vq->vq_nentries - 1));
-               uep = &vq->vq_ring.used->ring[used_idx];
+               uep = &vq->vq_split.ring.used->ring[used_idx];
                idx = (uint32_t) uep->id;
                desc_idx = idx;
 
-               while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
-                       desc_idx = vq->vq_ring.desc[desc_idx].next;
+               while (vq->vq_split.ring.desc[desc_idx].flags &
+                               VRING_DESC_F_NEXT) {
+                       desc_idx = vq->vq_split.ring.desc[desc_idx].next;
                        vq->vq_free_cnt++;
                }
 
-               vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
+               vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
                vq->vq_desc_head_idx = idx;
 
                vq->vq_used_cons_idx++;
@@ -244,6 +318,43 @@ virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
                        vq->vq_free_cnt, vq->vq_desc_head_idx);
 
        result = cvq->virtio_net_hdr_mz->addr;
+       return result;
+}
+
+static int
+virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
+                   int *dlen, int pkt_num)
+{
+       virtio_net_ctrl_ack status = ~0;
+       struct virtio_pmd_ctrl *result;
+       struct virtqueue *vq;
+
+       ctrl->status = status;
+
+       if (!cvq || !cvq->vq) {
+               PMD_INIT_LOG(ERR, "Control queue is not supported.");
+               return -1;
+       }
+
+       rte_spinlock_lock(&cvq->lock);
+       vq = cvq->vq;
+
+       PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
+               "vq->hw->cvq = %p vq = %p",
+               vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
+
+       if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
+               rte_spinlock_unlock(&cvq->lock);
+               return -1;
+       }
+
+       memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
+               sizeof(struct virtio_pmd_ctrl));
+
+       if (vtpci_packed_queue(vq->hw))
+               result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
+       else
+               result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
 
        rte_spinlock_unlock(&cvq->lock);
        return result->status;
@@ -294,25 +405,28 @@ static void
 virtio_init_vring(struct virtqueue *vq)
 {
        int size = vq->vq_nentries;
-       struct vring *vr = &vq->vq_ring;
        uint8_t *ring_mem = vq->vq_ring_virt_mem;
 
        PMD_INIT_FUNC_TRACE();
 
-       /*
-        * Reinitialise since virtio port might have been stopped and restarted
-        */
        memset(ring_mem, 0, vq->vq_ring_size);
-       vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
+
        vq->vq_used_cons_idx = 0;
        vq->vq_desc_head_idx = 0;
        vq->vq_avail_idx = 0;
        vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
        vq->vq_free_cnt = vq->vq_nentries;
        memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
+       if (vtpci_packed_queue(vq->hw)) {
+               vring_init_packed(&vq->vq_packed.ring, ring_mem,
+                                 VIRTIO_PCI_VRING_ALIGN, size);
+               vring_desc_init_packed(vq, size);
+       } else {
+               struct vring *vr = &vq->vq_split.ring;
 
-       vring_desc_init(vr->desc, size);
-
+               vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
+               vring_desc_init_split(vr->desc, size);
+       }
        /*
         * Disable device(host) interrupting guest
         */
@@ -335,8 +449,10 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
        void *sw_ring = NULL;
        int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
        int ret;
+       int numa_node = dev->device->numa_node;
 
-       PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
+       PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
+                       vtpci_queue_idx, numa_node);
 
        /*
         * Read the virtqueue size from the Queue Size field
@@ -349,8 +465,8 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
                return -EINVAL;
        }
 
-       if (!rte_is_power_of_2(vq_size)) {
-               PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
+       if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
+               PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
                return -EINVAL;
        }
 
@@ -372,7 +488,7 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
        }
 
        vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
-                               SOCKET_ID_ANY);
+                               numa_node);
        if (vq == NULL) {
                PMD_INIT_LOG(ERR, "can not allocate vq");
                return -ENOMEM;
@@ -382,18 +498,25 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
        vq->hw = hw;
        vq->vq_queue_index = vtpci_queue_idx;
        vq->vq_nentries = vq_size;
+       if (vtpci_packed_queue(hw)) {
+               vq->vq_packed.used_wrap_counter = 1;
+               vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
+               vq->vq_packed.event_flags_shadow = 0;
+               if (queue_type == VTNET_RQ)
+                       vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
+       }
 
        /*
         * Reserve a memzone for vring elements
         */
-       size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
+       size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
        vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
        PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
                     size, vq->vq_ring_size);
 
        mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
-                                        SOCKET_ID_ANY,
-                                        0, VIRTIO_PCI_VRING_ALIGN);
+                       numa_node, RTE_MEMZONE_IOVA_CONTIG,
+                       VIRTIO_PCI_VRING_ALIGN);
        if (mz == NULL) {
                if (rte_errno == EEXIST)
                        mz = rte_memzone_lookup(vq_name);
@@ -418,8 +541,8 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
                snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
                         dev->data->port_id, vtpci_queue_idx);
                hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
-                                                    SOCKET_ID_ANY, 0,
-                                                    RTE_CACHE_LINE_SIZE);
+                               numa_node, RTE_MEMZONE_IOVA_CONTIG,
+                               RTE_CACHE_LINE_SIZE);
                if (hdr_mz == NULL) {
                        if (rte_errno == EEXIST)
                                hdr_mz = rte_memzone_lookup(vq_hdr_name);
@@ -435,7 +558,7 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
                               sizeof(vq->sw_ring[0]);
 
                sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
-                               RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
+                               RTE_CACHE_LINE_SIZE, numa_node);
                if (!sw_ring) {
                        PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
                        ret = -ENOMEM;
@@ -465,8 +588,8 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
                hw->cvq = cvq;
        }
 
-       /* For virtio_user case (that is when hw->dev is NULL), we use
-        * virtual address. And we need properly set _offset_, please see
+       /* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
+        * we use virtual address. And we need properly set _offset_, please see
         * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
         */
        if (!hw->virtio_user_dev)
@@ -487,17 +610,28 @@ virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
                txr = hdr_mz->addr;
                memset(txr, 0, vq_size * sizeof(*txr));
                for (i = 0; i < vq_size; i++) {
-                       struct vring_desc *start_dp = txr[i].tx_indir;
-
-                       vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
-
                        /* first indirect descriptor is always the tx header */
-                       start_dp->addr = txvq->virtio_net_hdr_mem
-                               + i * sizeof(*txr)
-                               + offsetof(struct virtio_tx_region, tx_hdr);
-
-                       start_dp->len = hw->vtnet_hdr_size;
-                       start_dp->flags = VRING_DESC_F_NEXT;
+                       if (!vtpci_packed_queue(hw)) {
+                               struct vring_desc *start_dp = txr[i].tx_indir;
+                               vring_desc_init_split(start_dp,
+                                                     RTE_DIM(txr[i].tx_indir));
+                               start_dp->addr = txvq->virtio_net_hdr_mem
+                                       + i * sizeof(*txr)
+                                       + offsetof(struct virtio_tx_region,
+                                                  tx_hdr);
+                               start_dp->len = hw->vtnet_hdr_size;
+                               start_dp->flags = VRING_DESC_F_NEXT;
+                       } else {
+                               struct vring_packed_desc *start_dp =
+                                       txr[i].tx_packed_indir;
+                               vring_desc_init_indirect_packed(start_dp,
+                                     RTE_DIM(txr[i].tx_packed_indir));
+                               start_dp->addr = txvq->virtio_net_hdr_mem
+                                       + i * sizeof(*txr)
+                                       + offsetof(struct virtio_tx_region,
+                                                  tx_hdr);
+                               start_dp->len = hw->vtnet_hdr_size;
+                       }
                }
        }
 
@@ -580,13 +714,19 @@ virtio_alloc_queues(struct rte_eth_dev *dev)
 
 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 
-static void
+static int
 virtio_dev_close(struct rte_eth_dev *dev)
 {
        struct virtio_hw *hw = dev->data->dev_private;
        struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 
        PMD_INIT_LOG(DEBUG, "virtio_dev_close");
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return 0;
+
+       if (!hw->opened)
+               return 0;
+       hw->opened = false;
 
        /* reset the NIC */
        if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
@@ -604,9 +744,22 @@ virtio_dev_close(struct rte_eth_dev *dev)
        vtpci_reset(hw);
        virtio_dev_free_mbufs(dev);
        virtio_free_queues(hw);
+
+#ifdef RTE_VIRTIO_USER
+       if (hw->virtio_user_dev)
+               virtio_user_dev_uninit(hw->virtio_user_dev);
+       else
+#endif
+       if (dev->device) {
+               rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
+               if (!hw->modern)
+                       rte_pci_ioport_unmap(VTPCI_IO(hw));
+       }
+
+       return 0;
 }
 
-static void
+static int
 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 {
        struct virtio_hw *hw = dev->data->dev_private;
@@ -616,7 +769,7 @@ virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 
        if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
                PMD_INIT_LOG(INFO, "host does not support rx control");
-               return;
+               return -ENOTSUP;
        }
 
        ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
@@ -625,11 +778,15 @@ virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
        dlen[0] = 1;
 
        ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
-       if (ret)
+       if (ret) {
                PMD_INIT_LOG(ERR, "Failed to enable promisc");
+               return -EAGAIN;
+       }
+
+       return 0;
 }
 
-static void
+static int
 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 {
        struct virtio_hw *hw = dev->data->dev_private;
@@ -639,7 +796,7 @@ virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 
        if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
                PMD_INIT_LOG(INFO, "host does not support rx control");
-               return;
+               return -ENOTSUP;
        }
 
        ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
@@ -648,11 +805,15 @@ virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
        dlen[0] = 1;
 
        ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
-       if (ret)
+       if (ret) {
                PMD_INIT_LOG(ERR, "Failed to disable promisc");
+               return -EAGAIN;
+       }
+
+       return 0;
 }
 
-static void
+static int
 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 {
        struct virtio_hw *hw = dev->data->dev_private;
@@ -662,7 +823,7 @@ virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 
        if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
                PMD_INIT_LOG(INFO, "host does not support rx control");
-               return;
+               return -ENOTSUP;
        }
 
        ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
@@ -671,11 +832,15 @@ virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
        dlen[0] = 1;
 
        ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
-       if (ret)
+       if (ret) {
                PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
+               return -EAGAIN;
+       }
+
+       return 0;
 }
 
-static void
+static int
 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 {
        struct virtio_hw *hw = dev->data->dev_private;
@@ -685,7 +850,7 @@ virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 
        if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
                PMD_INIT_LOG(INFO, "host does not support rx control");
-               return;
+               return -ENOTSUP;
        }
 
        ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
@@ -694,8 +859,12 @@ virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
        dlen[0] = 1;
 
        ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
-       if (ret)
+       if (ret) {
                PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
+               return -EAGAIN;
+       }
+
+       return 0;
 }
 
 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
@@ -703,16 +872,16 @@ static int
 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 {
        struct virtio_hw *hw = dev->data->dev_private;
-       uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
+       uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
                                 hw->vtnet_hdr_size;
        uint32_t frame_size = mtu + ether_hdr_len;
        uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 
        max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 
-       if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
+       if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
                PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
-                       ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
+                       RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
                return -EINVAL;
        }
        return 0;
@@ -721,10 +890,12 @@ virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 static int
 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 {
+       struct virtio_hw *hw = dev->data->dev_private;
        struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
        struct virtqueue *vq = rxvq->vq;
 
        virtqueue_enable_intr(vq);
+       virtio_mb(hw->weak_barriers);
        return 0;
 }
 
@@ -763,7 +934,6 @@ static const struct eth_dev_ops virtio_eth_dev_ops = {
        .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
        .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
        .rx_queue_release        = virtio_dev_queue_release,
-       .rx_descriptor_done      = virtio_dev_rx_queue_done,
        .tx_queue_setup          = virtio_dev_tx_queue_setup,
        .tx_queue_release        = virtio_dev_queue_release,
        /* collect stats per queue */
@@ -774,45 +944,20 @@ static const struct eth_dev_ops virtio_eth_dev_ops = {
        .mac_addr_set            = virtio_mac_addr_set,
 };
 
-static inline int
-virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
-                               struct rte_eth_link *link)
-{
-       struct rte_eth_link *dst = link;
-       struct rte_eth_link *src = &(dev->data->dev_link);
-
-       if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
-                       *(uint64_t *)src) == 0)
-               return -1;
-
-       return 0;
-}
-
-/**
- * Atomically writes the link status information into global
- * structure rte_eth_dev.
- *
- * @param dev
- *   - Pointer to the structure rte_eth_dev to read from.
- *   - Pointer to the buffer to be saved with the link status.
- *
- * @return
- *   - On success, zero.
- *   - On failure, negative value.
+/*
+ * dev_ops for virtio-user in secondary processes, as we just have
+ * some limited supports currently.
  */
-static inline int
-virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
-               struct rte_eth_link *link)
-{
-       struct rte_eth_link *dst = &(dev->data->dev_link);
-       struct rte_eth_link *src = link;
-
-       if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
-                                       *(uint64_t *)src) == 0)
-               return -1;
-
-       return 0;
-}
+const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
+       .dev_infos_get           = virtio_dev_info_get,
+       .stats_get               = virtio_dev_stats_get,
+       .xstats_get              = virtio_dev_xstats_get,
+       .xstats_get_names        = virtio_dev_xstats_get_names,
+       .stats_reset             = virtio_dev_stats_reset,
+       .xstats_reset            = virtio_dev_stats_reset,
+       /* collect stats per queue */
+       .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
+};
 
 static void
 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
@@ -826,7 +971,6 @@ virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 
                stats->opackets += txvq->stats.packets;
                stats->obytes += txvq->stats.bytes;
-               stats->oerrors += txvq->stats.errors;
 
                if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
                        stats->q_opackets[i] = txvq->stats.packets;
@@ -952,7 +1096,7 @@ virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
        return 0;
 }
 
-static void
+static int
 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 {
        unsigned int i;
@@ -964,7 +1108,6 @@ virtio_dev_stats_reset(struct rte_eth_dev *dev)
 
                txvq->stats.packets = 0;
                txvq->stats.bytes = 0;
-               txvq->stats.errors = 0;
                txvq->stats.multicast = 0;
                txvq->stats.broadcast = 0;
                memset(txvq->stats.size_bins, 0,
@@ -984,6 +1127,8 @@ virtio_dev_stats_reset(struct rte_eth_dev *dev)
                memset(rxvq->stats.size_bins, 0,
                       sizeof(rxvq->stats.size_bins[0]) * 8);
        }
+
+       return 0;
 }
 
 static void
@@ -991,7 +1136,7 @@ virtio_set_hwaddr(struct virtio_hw *hw)
 {
        vtpci_write_dev_config(hw,
                        offsetof(struct virtio_net_config, mac),
-                       &hw->mac_addr, ETHER_ADDR_LEN);
+                       &hw->mac_addr, RTE_ETHER_ADDR_LEN);
 }
 
 static void
@@ -1000,9 +1145,9 @@ virtio_get_hwaddr(struct virtio_hw *hw)
        if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
                vtpci_read_dev_config(hw,
                        offsetof(struct virtio_net_config, mac),
-                       &hw->mac_addr, ETHER_ADDR_LEN);
+                       &hw->mac_addr, RTE_ETHER_ADDR_LEN);
        } else {
-               eth_random_addr(&hw->mac_addr[0]);
+               rte_eth_random_addr(&hw->mac_addr[0]);
                virtio_set_hwaddr(hw);
        }
 }
@@ -1023,10 +1168,10 @@ virtio_mac_table_set(struct virtio_hw *hw,
        ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
        ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 
-       len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
+       len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
        memcpy(ctrl.data, uc, len[0]);
 
-       len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
+       len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
        memcpy(ctrl.data + len[0], mc, len[1]);
 
        err = virtio_send_command(hw->cvq, &ctrl, len, 2);
@@ -1036,11 +1181,11 @@ virtio_mac_table_set(struct virtio_hw *hw,
 }
 
 static int
-virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
+virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
                    uint32_t index, uint32_t vmdq __rte_unused)
 {
        struct virtio_hw *hw = dev->data->dev_private;
-       const struct ether_addr *addrs = dev->data->mac_addrs;
+       const struct rte_ether_addr *addrs = dev->data->mac_addrs;
        unsigned int i;
        struct virtio_net_ctrl_mac *uc, *mc;
 
@@ -1049,18 +1194,20 @@ virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
                return -EINVAL;
        }
 
-       uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
+       uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
+               sizeof(uc->entries));
        uc->entries = 0;
-       mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
+       mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
+               sizeof(mc->entries));
        mc->entries = 0;
 
        for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
-               const struct ether_addr *addr
+               const struct rte_ether_addr *addr
                        = (i == index) ? mac_addr : addrs + i;
                struct virtio_net_ctrl_mac *tbl
-                       = is_multicast_ether_addr(addr) ? mc : uc;
+                       = rte_is_multicast_ether_addr(addr) ? mc : uc;
 
-               memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
+               memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
        }
 
        return virtio_mac_table_set(hw, uc, mc);
@@ -1070,7 +1217,7 @@ static void
 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 {
        struct virtio_hw *hw = dev->data->dev_private;
-       struct ether_addr *addrs = dev->data->mac_addrs;
+       struct rte_ether_addr *addrs = dev->data->mac_addrs;
        struct virtio_net_ctrl_mac *uc, *mc;
        unsigned int i;
 
@@ -1079,43 +1226,51 @@ virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
                return;
        }
 
-       uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
+       uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
+               sizeof(uc->entries));
        uc->entries = 0;
-       mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
+       mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
+               sizeof(mc->entries));
        mc->entries = 0;
 
        for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
                struct virtio_net_ctrl_mac *tbl;
 
-               if (i == index || is_zero_ether_addr(addrs + i))
+               if (i == index || rte_is_zero_ether_addr(addrs + i))
                        continue;
 
-               tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
-               memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
+               tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
+               memcpy(&tbl->macs[tbl->entries++], addrs + i,
+                       RTE_ETHER_ADDR_LEN);
        }
 
        virtio_mac_table_set(hw, uc, mc);
 }
 
-static void
-virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
+static int
+virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
 {
        struct virtio_hw *hw = dev->data->dev_private;
 
-       memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
+       memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
 
        /* Use atomic update if available */
        if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
                struct virtio_pmd_ctrl ctrl;
-               int len = ETHER_ADDR_LEN;
+               int len = RTE_ETHER_ADDR_LEN;
 
                ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
                ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 
-               memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
-               virtio_send_command(hw->cvq, &ctrl, &len, 1);
-       } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
-               virtio_set_hwaddr(hw);
+               memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
+               return virtio_send_command(hw->cvq, &ctrl, &len, 1);
+       }
+
+       if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
+               return -ENOTSUP;
+
+       virtio_set_hwaddr(hw);
+       return 0;
 }
 
 static int
@@ -1136,6 +1291,20 @@ virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
        return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 }
 
+static int
+virtio_intr_unmask(struct rte_eth_dev *dev)
+{
+       struct virtio_hw *hw = dev->data->dev_private;
+
+       if (rte_intr_ack(dev->intr_handle) < 0)
+               return -1;
+
+       if (!hw->virtio_user_dev)
+               hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
+
+       return 0;
+}
+
 static int
 virtio_intr_enable(struct rte_eth_dev *dev)
 {
@@ -1186,7 +1355,7 @@ virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
                        offsetof(struct virtio_net_config, mtu),
                        &config.mtu, sizeof(config.mtu));
 
-               if (config.mtu < ETHER_MIN_MTU)
+               if (config.mtu < RTE_ETHER_MIN_MTU)
                        req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
        }
 
@@ -1199,12 +1368,13 @@ virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
        PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
                hw->guest_features);
 
-       if (hw->modern) {
-               if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
-                       PMD_INIT_LOG(ERR,
-                               "VIRTIO_F_VERSION_1 features is not enabled.");
-                       return -1;
-               }
+       if (hw->modern && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
+               PMD_INIT_LOG(ERR,
+                       "VIRTIO_F_VERSION_1 features is not enabled.");
+               return -1;
+       }
+
+       if (hw->modern || hw->virtio_user_dev) {
                vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
                if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
                        PMD_INIT_LOG(ERR,
@@ -1273,11 +1443,18 @@ static void
 virtio_notify_peers(struct rte_eth_dev *dev)
 {
        struct virtio_hw *hw = dev->data->dev_private;
-       struct virtnet_rx *rxvq = dev->data->rx_queues[0];
+       struct virtnet_rx *rxvq;
        struct rte_mbuf *rarp_mbuf;
 
+       if (!dev->data->rx_queues)
+               return;
+
+       rxvq = dev->data->rx_queues[0];
+       if (!rxvq)
+               return;
+
        rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
-                       (struct ether_addr *)hw->mac_addr);
+                       (struct rte_ether_addr *)hw->mac_addr);
        if (rarp_mbuf == NULL) {
                PMD_DRV_LOG(ERR, "failed to make RARP packet.");
                return;
@@ -1316,24 +1493,31 @@ virtio_interrupt_handler(void *param)
        struct rte_eth_dev *dev = param;
        struct virtio_hw *hw = dev->data->dev_private;
        uint8_t isr;
+       uint16_t status;
 
        /* Read interrupt status which clears interrupt */
        isr = vtpci_isr(hw);
        PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
 
-       if (virtio_intr_enable(dev) < 0)
+       if (virtio_intr_unmask(dev) < 0)
                PMD_DRV_LOG(ERR, "interrupt enable failed");
 
        if (isr & VIRTIO_PCI_ISR_CONFIG) {
                if (virtio_dev_link_update(dev, 0) == 0)
-                       _rte_eth_dev_callback_process(dev,
-                                                     RTE_ETH_EVENT_INTR_LSC,
-                                                     NULL);
-       }
+                       rte_eth_dev_callback_process(dev,
+                                                    RTE_ETH_EVENT_INTR_LSC,
+                                                    NULL);
 
-       if (isr & VIRTIO_NET_S_ANNOUNCE) {
-               virtio_notify_peers(dev);
-               virtio_ack_link_announce(dev);
+               if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
+                       vtpci_read_dev_config(hw,
+                               offsetof(struct virtio_net_config, status),
+                               &status, sizeof(status));
+                       if (status & VIRTIO_NET_S_ANNOUNCE) {
+                               virtio_notify_peers(dev);
+                               if (hw->cvq)
+                                       virtio_ack_link_announce(dev);
+                       }
+               }
        }
 }
 
@@ -1343,30 +1527,69 @@ set_rxtx_funcs(struct rte_eth_dev *eth_dev)
 {
        struct virtio_hw *hw = eth_dev->data->dev_private;
 
-       if (hw->use_simple_rx) {
-               PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
-                       eth_dev->data->port_id);
-               eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
-       } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
+       eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
+       if (vtpci_packed_queue(hw)) {
                PMD_INIT_LOG(INFO,
-                       "virtio: using mergeable buffer Rx path on port %u",
+                       "virtio: using packed ring %s Tx path on port %u",
+                       hw->use_vec_tx ? "vectorized" : "standard",
                        eth_dev->data->port_id);
-               eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
+               if (hw->use_vec_tx)
+                       eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
+               else
+                       eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
        } else {
-               PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
-                       eth_dev->data->port_id);
-               eth_dev->rx_pkt_burst = &virtio_recv_pkts;
+               if (hw->use_inorder_tx) {
+                       PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
+               } else {
+                       PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->tx_pkt_burst = virtio_xmit_pkts;
+               }
        }
 
-       if (hw->use_simple_tx) {
-               PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
-                       eth_dev->data->port_id);
-               eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
+       if (vtpci_packed_queue(hw)) {
+               if (hw->use_vec_rx) {
+                       PMD_INIT_LOG(INFO,
+                               "virtio: using packed ring vectorized Rx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->rx_pkt_burst =
+                               &virtio_recv_pkts_packed_vec;
+               } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
+                       PMD_INIT_LOG(INFO,
+                               "virtio: using packed ring mergeable buffer Rx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->rx_pkt_burst =
+                               &virtio_recv_mergeable_pkts_packed;
+               } else {
+                       PMD_INIT_LOG(INFO,
+                               "virtio: using packed ring standard Rx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
+               }
        } else {
-               PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
-                       eth_dev->data->port_id);
-               eth_dev->tx_pkt_burst = virtio_xmit_pkts;
+               if (hw->use_vec_rx) {
+                       PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
+               } else if (hw->use_inorder_rx) {
+                       PMD_INIT_LOG(INFO,
+                               "virtio: using inorder Rx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
+               } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
+                       PMD_INIT_LOG(INFO,
+                               "virtio: using mergeable buffer Rx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
+               } else {
+                       PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
+                               eth_dev->data->port_id);
+                       eth_dev->rx_pkt_burst = &virtio_recv_pkts;
+               }
        }
+
 }
 
 /* Only support 1:1 queue/interrupt mapping so far.
@@ -1457,7 +1680,7 @@ virtio_configure_intr(struct rte_eth_dev *dev)
 
        return 0;
 }
-
+#define DUPLEX_UNKNOWN   0xff
 /* reset device and renegotiate features if needed */
 static int
 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
@@ -1471,6 +1694,11 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
        /* Reset the device although not necessary at startup */
        vtpci_reset(hw);
 
+       if (hw->vqs) {
+               virtio_dev_free_mbufs(eth_dev);
+               virtio_free_queues(hw);
+       }
+
        /* Tell the host we've noticed this device. */
        vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
 
@@ -1479,10 +1707,10 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
        if (virtio_negotiate_features(hw, req_features) < 0)
                return -1;
 
-       if (!hw->virtio_user_dev) {
+       hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
+
+       if (!hw->virtio_user_dev)
                pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
-               rte_eth_copy_pci_info(eth_dev, pci_dev);
-       }
 
        /* If host does not support both status and MSI-X then disable LSC */
        if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
@@ -1491,22 +1719,42 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
        else
                eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
 
+       eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
+
        /* Setting up rx_header size for the device */
        if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
-           vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
+           vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
+           vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
                hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
        else
                hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
 
        /* Copy the permanent MAC address to: virtio_hw */
        virtio_get_hwaddr(hw);
-       ether_addr_copy((struct ether_addr *) hw->mac_addr,
+       rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
                        &eth_dev->data->mac_addrs[0]);
        PMD_INIT_LOG(DEBUG,
                     "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
                     hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
                     hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
 
+       if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
+               if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
+                       config = &local_config;
+                       vtpci_read_dev_config(hw,
+                               offsetof(struct virtio_net_config, speed),
+                               &config->speed, sizeof(config->speed));
+                       vtpci_read_dev_config(hw,
+                               offsetof(struct virtio_net_config, duplex),
+                               &config->duplex, sizeof(config->duplex));
+                       hw->speed = config->speed;
+                       hw->duplex = config->duplex;
+               }
+       }
+       if (hw->duplex == DUPLEX_UNKNOWN)
+               hw->duplex = ETH_LINK_FULL_DUPLEX;
+       PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
+               hw->speed, hw->duplex);
        if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
                config = &local_config;
 
@@ -1548,7 +1796,7 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
                         * time, but check again in case it has changed since
                         * then, which should not happen.
                         */
-                       if (config->mtu < ETHER_MIN_MTU) {
+                       if (config->mtu < RTE_ETHER_MIN_MTU) {
                                PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
                                                config->mtu);
                                return -1;
@@ -1559,7 +1807,7 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
                        eth_dev->data->mtu = config->mtu;
 
                } else {
-                       hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
+                       hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
                                VLAN_TAG_LEN - hw->vtnet_hdr_size;
                }
 
@@ -1574,7 +1822,7 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
        } else {
                PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
                hw->max_queue_pairs = 1;
-               hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
+               hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
                        VLAN_TAG_LEN - hw->vtnet_hdr_size;
        }
 
@@ -1585,6 +1833,7 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
        if (eth_dev->data->dev_conf.intr_conf.rxq) {
                if (virtio_configure_intr(eth_dev) < 0) {
                        PMD_INIT_LOG(ERR, "failed to configure interrupt");
+                       virtio_free_queues(hw);
                        return -1;
                }
        }
@@ -1653,11 +1902,21 @@ int
 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
 {
        struct virtio_hw *hw = eth_dev->data->dev_private;
+       uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
+       int vectorized = 0;
        int ret;
 
-       RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
+       if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
+               PMD_INIT_LOG(ERR,
+                       "Not sufficient headroom required = %d, avail = %d",
+                       (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
+                       RTE_PKTMBUF_HEADROOM);
+
+               return -1;
+       }
 
        eth_dev->dev_ops = &virtio_eth_dev_ops;
+       eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
 
        if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
                if (!hw->virtio_user_dev) {
@@ -1671,13 +1930,19 @@ eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
 
                return 0;
        }
+       ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
+                NULL, &speed, &vectorized);
+       if (ret < 0)
+               return ret;
+       hw->speed = speed;
 
        /* Allocate memory for storing MAC addresses */
-       eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
+       eth_dev->data->mac_addrs = rte_zmalloc("virtio",
+                               VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
        if (eth_dev->data->mac_addrs == NULL) {
                PMD_INIT_LOG(ERR,
                        "Failed to allocate %d bytes needed to store MAC addresses",
-                       VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
+                       VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
                return -ENOMEM;
        }
 
@@ -1688,67 +1953,218 @@ eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
        if (!hw->virtio_user_dev) {
                ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
                if (ret)
-                       goto out;
+                       goto err_vtpci_init;
        }
 
+       rte_spinlock_init(&hw->state_lock);
+
        /* reset device and negotiate default features */
        ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
        if (ret < 0)
-               goto out;
+               goto err_virtio_init;
+
+       if (vectorized) {
+               if (!vtpci_packed_queue(hw)) {
+                       hw->use_vec_rx = 1;
+               } else {
+#if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
+                       hw->use_vec_rx = 1;
+                       hw->use_vec_tx = 1;
+#else
+                       PMD_DRV_LOG(INFO,
+                               "building environment do not support packed ring vectorized");
+#endif
+               }
+       }
 
-       /* Setup interrupt callback  */
-       if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
-               rte_intr_callback_register(eth_dev->intr_handle,
-                       virtio_interrupt_handler, eth_dev);
+       hw->opened = true;
 
        return 0;
 
-out:
+err_virtio_init:
+       if (!hw->virtio_user_dev) {
+               rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
+               if (!hw->modern)
+                       rte_pci_ioport_unmap(VTPCI_IO(hw));
+       }
+err_vtpci_init:
        rte_free(eth_dev->data->mac_addrs);
+       eth_dev->data->mac_addrs = NULL;
        return ret;
 }
 
 static int
 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
 {
+       int ret;
        PMD_INIT_FUNC_TRACE();
 
        if (rte_eal_process_type() == RTE_PROC_SECONDARY)
-               return -EPERM;
+               return 0;
 
-       virtio_dev_stop(eth_dev);
+       ret = virtio_dev_stop(eth_dev);
        virtio_dev_close(eth_dev);
 
-       eth_dev->dev_ops = NULL;
-       eth_dev->tx_pkt_burst = NULL;
-       eth_dev->rx_pkt_burst = NULL;
+       PMD_INIT_LOG(DEBUG, "dev_uninit completed");
 
-       rte_free(eth_dev->data->mac_addrs);
-       eth_dev->data->mac_addrs = NULL;
+       return ret;
+}
 
-       /* reset interrupt callback  */
-       if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
-               rte_intr_callback_unregister(eth_dev->intr_handle,
-                                               virtio_interrupt_handler,
-                                               eth_dev);
-       if (eth_dev->device)
-               rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
 
-       PMD_INIT_LOG(DEBUG, "dev_uninit completed");
+static int vdpa_check_handler(__rte_unused const char *key,
+               const char *value, void *ret_val)
+{
+       if (strcmp(value, "1") == 0)
+               *(int *)ret_val = 1;
+       else
+               *(int *)ret_val = 0;
+
+       return 0;
+}
+
+
+static uint32_t
+virtio_dev_speed_capa_get(uint32_t speed)
+{
+       switch (speed) {
+       case ETH_SPEED_NUM_10G:
+               return ETH_LINK_SPEED_10G;
+       case ETH_SPEED_NUM_20G:
+               return ETH_LINK_SPEED_20G;
+       case ETH_SPEED_NUM_25G:
+               return ETH_LINK_SPEED_25G;
+       case ETH_SPEED_NUM_40G:
+               return ETH_LINK_SPEED_40G;
+       case ETH_SPEED_NUM_50G:
+               return ETH_LINK_SPEED_50G;
+       case ETH_SPEED_NUM_56G:
+               return ETH_LINK_SPEED_56G;
+       case ETH_SPEED_NUM_100G:
+               return ETH_LINK_SPEED_100G;
+       case ETH_SPEED_NUM_200G:
+               return ETH_LINK_SPEED_200G;
+       default:
+               return 0;
+       }
+}
+
+static int vectorized_check_handler(__rte_unused const char *key,
+               const char *value, void *ret_val)
+{
+       if (strcmp(value, "1") == 0)
+               *(int *)ret_val = 1;
+       else
+               *(int *)ret_val = 0;
 
        return 0;
 }
 
+#define VIRTIO_ARG_SPEED      "speed"
+#define VIRTIO_ARG_VDPA       "vdpa"
+#define VIRTIO_ARG_VECTORIZED "vectorized"
+
+
+static int
+link_speed_handler(const char *key __rte_unused,
+               const char *value, void *ret_val)
+{
+       uint32_t val;
+       if (!value || !ret_val)
+               return -EINVAL;
+       val = strtoul(value, NULL, 0);
+       /* validate input */
+       if (virtio_dev_speed_capa_get(val) == 0)
+               return -EINVAL;
+       *(uint32_t *)ret_val = val;
+
+       return 0;
+}
+
+
+static int
+virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
+       uint32_t *speed, int *vectorized)
+{
+       struct rte_kvargs *kvlist;
+       int ret = 0;
+
+       if (devargs == NULL)
+               return 0;
+
+       kvlist = rte_kvargs_parse(devargs->args, NULL);
+       if (kvlist == NULL) {
+               PMD_INIT_LOG(ERR, "error when parsing param");
+               return 0;
+       }
+       if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
+               /* vdpa mode selected when there's a key-value pair:
+                * vdpa=1
+                */
+               ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
+                               vdpa_check_handler, vdpa);
+               if (ret < 0) {
+                       PMD_INIT_LOG(ERR, "Failed to parse %s",
+                               VIRTIO_ARG_VDPA);
+                       goto exit;
+               }
+       }
+       if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
+               ret = rte_kvargs_process(kvlist,
+                                       VIRTIO_ARG_SPEED,
+                                       link_speed_handler, speed);
+               if (ret < 0) {
+                       PMD_INIT_LOG(ERR, "Failed to parse %s",
+                                       VIRTIO_ARG_SPEED);
+                       goto exit;
+               }
+       }
+
+       if (vectorized &&
+               rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
+               ret = rte_kvargs_process(kvlist,
+                               VIRTIO_ARG_VECTORIZED,
+                               vectorized_check_handler, vectorized);
+               if (ret < 0) {
+                       PMD_INIT_LOG(ERR, "Failed to parse %s",
+                                       VIRTIO_ARG_VECTORIZED);
+                       goto exit;
+               }
+       }
+
+exit:
+       rte_kvargs_free(kvlist);
+       return ret;
+}
+
 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
        struct rte_pci_device *pci_dev)
 {
+       int vdpa = 0;
+       int ret = 0;
+
+       ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
+               NULL);
+       if (ret < 0) {
+               PMD_INIT_LOG(ERR, "devargs parsing is failed");
+               return ret;
+       }
+       /* virtio pmd skips probe if device needs to work in vdpa mode */
+       if (vdpa == 1)
+               return 1;
+
        return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
                eth_virtio_dev_init);
 }
 
 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
 {
-       return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
+       int ret;
+
+       ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
+       /* Port has already been released by close. */
+       if (ret == -ENODEV)
+               ret = 0;
+       return ret;
 }
 
 static struct rte_pci_driver rte_virtio_pmd = {
@@ -1761,18 +2177,28 @@ static struct rte_pci_driver rte_virtio_pmd = {
        .remove = eth_virtio_pci_remove,
 };
 
-RTE_INIT(rte_virtio_pmd_init);
-static void
-rte_virtio_pmd_init(void)
+RTE_INIT(rte_virtio_pmd_init)
 {
-       if (rte_eal_iopl_init() != 0) {
-               PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
-               return;
-       }
-
+       rte_eal_iopl_init();
        rte_pci_register(&rte_virtio_pmd);
 }
 
+static bool
+rx_offload_enabled(struct virtio_hw *hw)
+{
+       return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
+               vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
+               vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
+}
+
+static bool
+tx_offload_enabled(struct virtio_hw *hw)
+{
+       return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
+               vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
+               vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
+}
+
 /*
  * Configure virtio device
  * It returns 0 on success.
@@ -1781,31 +2207,59 @@ static int
 virtio_dev_configure(struct rte_eth_dev *dev)
 {
        const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
+       const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
        struct virtio_hw *hw = dev->data->dev_private;
+       uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
+               hw->vtnet_hdr_size;
+       uint64_t rx_offloads = rxmode->offloads;
+       uint64_t tx_offloads = txmode->offloads;
        uint64_t req_features;
        int ret;
 
        PMD_INIT_LOG(DEBUG, "configure");
        req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
 
+       if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
+               PMD_DRV_LOG(ERR,
+                       "Unsupported Rx multi queue mode %d",
+                       rxmode->mq_mode);
+               return -EINVAL;
+       }
+
+       if (txmode->mq_mode != ETH_MQ_TX_NONE) {
+               PMD_DRV_LOG(ERR,
+                       "Unsupported Tx multi queue mode %d",
+                       txmode->mq_mode);
+               return -EINVAL;
+       }
+
        if (dev->data->dev_conf.intr_conf.rxq) {
                ret = virtio_init_device(dev, hw->req_guest_features);
                if (ret < 0)
                        return ret;
        }
 
-       /* The name hw_ip_checksum is a bit confusing since it can be
-        * set by the application to request L3 and/or L4 checksums. In
-        * case of virtio, only L4 checksum is supported.
-        */
-       if (rxmode->hw_ip_checksum)
+       if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
+               req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
+
+       if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
+                          DEV_RX_OFFLOAD_TCP_CKSUM))
                req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
 
-       if (rxmode->enable_lro)
+       if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
                req_features |=
                        (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
                        (1ULL << VIRTIO_NET_F_GUEST_TSO6);
 
+       if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
+                          DEV_TX_OFFLOAD_TCP_CKSUM))
+               req_features |= (1ULL << VIRTIO_NET_F_CSUM);
+
+       if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
+               req_features |=
+                       (1ULL << VIRTIO_NET_F_HOST_TSO4) |
+                       (1ULL << VIRTIO_NET_F_HOST_TSO6);
+
        /* if request features changed, reinit the device */
        if (req_features != hw->req_guest_features) {
                ret = virtio_init_device(dev, req_features);
@@ -1813,14 +2267,15 @@ virtio_dev_configure(struct rte_eth_dev *dev)
                        return ret;
        }
 
-       if (rxmode->hw_ip_checksum &&
+       if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
+                           DEV_RX_OFFLOAD_TCP_CKSUM)) &&
                !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
                PMD_DRV_LOG(ERR,
                        "rx checksum not available on this host");
                return -ENOTSUP;
        }
 
-       if (rxmode->enable_lro &&
+       if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
                (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
                PMD_DRV_LOG(ERR,
@@ -1832,15 +2287,19 @@ virtio_dev_configure(struct rte_eth_dev *dev)
        if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
                virtio_dev_cq_start(dev);
 
-       hw->vlan_strip = rxmode->hw_vlan_strip;
+       if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
+               hw->vlan_strip = 1;
 
-       if (rxmode->hw_vlan_filter
+       if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
            && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
                PMD_DRV_LOG(ERR,
                            "vlan filtering not available on this host");
                return -ENOTSUP;
        }
 
+       hw->has_tx_offload = tx_offload_enabled(hw);
+       hw->has_rx_offload = rx_offload_enabled(hw);
+
        if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
                /* Enable vector (0) for Link State Intrerrupt */
                if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
@@ -1849,24 +2308,84 @@ virtio_dev_configure(struct rte_eth_dev *dev)
                        return -EBUSY;
                }
 
-       rte_spinlock_init(&hw->state_lock);
+       if (vtpci_packed_queue(hw)) {
+#if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
+               if ((hw->use_vec_rx || hw->use_vec_tx) &&
+                   (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
+                    !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
+                    !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
+                    rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
+                       PMD_DRV_LOG(INFO,
+                               "disabled packed ring vectorized path for requirements not met");
+                       hw->use_vec_rx = 0;
+                       hw->use_vec_tx = 0;
+               }
+#elif defined(RTE_ARCH_ARM)
+               if ((hw->use_vec_rx || hw->use_vec_tx) &&
+                   (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
+                    !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
+                    !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
+                    rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
+                       PMD_DRV_LOG(INFO,
+                               "disabled packed ring vectorized path for requirements not met");
+                       hw->use_vec_rx = 0;
+                       hw->use_vec_tx = 0;
+               }
+#else
+               hw->use_vec_rx = 0;
+               hw->use_vec_tx = 0;
+#endif
 
-       hw->use_simple_rx = 1;
-       hw->use_simple_tx = 1;
+               if (hw->use_vec_rx) {
+                       if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
+                               PMD_DRV_LOG(INFO,
+                                       "disabled packed ring vectorized rx for mrg_rxbuf enabled");
+                               hw->use_vec_rx = 0;
+                       }
 
-#if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
-       if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
-               hw->use_simple_rx = 0;
-               hw->use_simple_tx = 0;
-       }
+                       if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
+                               PMD_DRV_LOG(INFO,
+                                       "disabled packed ring vectorized rx for TCP_LRO enabled");
+                               hw->use_vec_rx = 0;
+                       }
+               }
+       } else {
+               if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
+                       hw->use_inorder_tx = 1;
+                       hw->use_inorder_rx = 1;
+                       hw->use_vec_rx = 0;
+               }
+
+               if (hw->use_vec_rx) {
+#if defined RTE_ARCH_ARM
+                       if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
+                               PMD_DRV_LOG(INFO,
+                                       "disabled split ring vectorized path for requirement not met");
+                               hw->use_vec_rx = 0;
+                       }
 #endif
-       if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
-               hw->use_simple_rx = 0;
-               hw->use_simple_tx = 0;
-       }
+                       if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
+                               PMD_DRV_LOG(INFO,
+                                       "disabled split ring vectorized rx for mrg_rxbuf enabled");
+                               hw->use_vec_rx = 0;
+                       }
 
-       if (rxmode->hw_ip_checksum)
-               hw->use_simple_rx = 0;
+                       if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
+                                          DEV_RX_OFFLOAD_TCP_CKSUM |
+                                          DEV_RX_OFFLOAD_TCP_LRO |
+                                          DEV_RX_OFFLOAD_VLAN_STRIP)) {
+                               PMD_DRV_LOG(INFO,
+                                       "disabled split ring vectorized rx for offloading enabled");
+                               hw->use_vec_rx = 0;
+                       }
+
+                       if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
+                               PMD_DRV_LOG(INFO,
+                                       "disabled split ring vectorized rx, max SIMD bitwidth too low");
+                               hw->use_vec_rx = 0;
+                       }
+               }
+       }
 
        return 0;
 }
@@ -1909,6 +2428,12 @@ virtio_dev_start(struct rte_eth_dev *dev)
            dev->data->dev_conf.intr_conf.rxq) {
                virtio_intr_disable(dev);
 
+               /* Setup interrupt callback  */
+               if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
+                       rte_intr_callback_register(dev->intr_handle,
+                                                  virtio_interrupt_handler,
+                                                  dev);
+
                if (virtio_intr_enable(dev) < 0) {
                        PMD_DRV_LOG(ERR, "interrupt enable failed");
                        return -EIO;
@@ -1952,7 +2477,7 @@ virtio_dev_start(struct rte_eth_dev *dev)
        }
 
        set_rxtx_funcs(dev);
-       hw->started = 1;
+       hw->started = true;
 
        /* Initialize Link state */
        virtio_dev_link_update(dev, 0);
@@ -1962,53 +2487,53 @@ virtio_dev_start(struct rte_eth_dev *dev)
 
 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
 {
+       struct virtio_hw *hw = dev->data->dev_private;
+       uint16_t nr_vq = virtio_get_nr_vq(hw);
+       const char *type __rte_unused;
+       unsigned int i, mbuf_num = 0;
+       struct virtqueue *vq;
        struct rte_mbuf *buf;
-       int i, mbuf_num = 0;
-
-       for (i = 0; i < dev->data->nb_rx_queues; i++) {
-               struct virtnet_rx *rxvq = dev->data->rx_queues[i];
-
-               PMD_INIT_LOG(DEBUG,
-                            "Before freeing rxq[%d] used and unused buf", i);
-               VIRTQUEUE_DUMP(rxvq->vq);
+       int queue_type;
 
-               PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
-               while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
-                       rte_pktmbuf_free(buf);
-                       mbuf_num++;
-               }
+       if (hw->vqs == NULL)
+               return;
 
-               PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
-               PMD_INIT_LOG(DEBUG,
-                            "After freeing rxq[%d] used and unused buf", i);
-               VIRTQUEUE_DUMP(rxvq->vq);
-       }
+       for (i = 0; i < nr_vq; i++) {
+               vq = hw->vqs[i];
+               if (!vq)
+                       continue;
 
-       for (i = 0; i < dev->data->nb_tx_queues; i++) {
-               struct virtnet_tx *txvq = dev->data->tx_queues[i];
+               queue_type = virtio_get_queue_type(hw, i);
+               if (queue_type == VTNET_RQ)
+                       type = "rxq";
+               else if (queue_type == VTNET_TQ)
+                       type = "txq";
+               else
+                       continue;
 
                PMD_INIT_LOG(DEBUG,
-                            "Before freeing txq[%d] used and unused bufs",
-                            i);
-               VIRTQUEUE_DUMP(txvq->vq);
+                       "Before freeing %s[%d] used and unused buf",
+                       type, i);
+               VIRTQUEUE_DUMP(vq);
 
-               mbuf_num = 0;
-               while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
+               while ((buf = virtqueue_detach_unused(vq)) != NULL) {
                        rte_pktmbuf_free(buf);
                        mbuf_num++;
                }
 
-               PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
                PMD_INIT_LOG(DEBUG,
-                            "After freeing txq[%d] used and unused buf", i);
-               VIRTQUEUE_DUMP(txvq->vq);
+                       "After freeing %s[%d] used and unused buf",
+                       type, i);
+               VIRTQUEUE_DUMP(vq);
        }
+
+       PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
 }
 
 /*
  * Stop device: disable interrupt and mark link down
  */
-static void
+static int
 virtio_dev_stop(struct rte_eth_dev *dev)
 {
        struct virtio_hw *hw = dev->data->dev_private;
@@ -2016,31 +2541,47 @@ virtio_dev_stop(struct rte_eth_dev *dev)
        struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 
        PMD_INIT_LOG(DEBUG, "stop");
+       dev->data->dev_started = 0;
 
        rte_spinlock_lock(&hw->state_lock);
-       if (intr_conf->lsc || intr_conf->rxq)
+       if (!hw->started)
+               goto out_unlock;
+       hw->started = false;
+
+       if (intr_conf->lsc || intr_conf->rxq) {
                virtio_intr_disable(dev);
 
-       hw->started = 0;
+               /* Reset interrupt callback  */
+               if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
+                       rte_intr_callback_unregister(dev->intr_handle,
+                                                    virtio_interrupt_handler,
+                                                    dev);
+               }
+       }
+
        memset(&link, 0, sizeof(link));
-       virtio_dev_atomic_write_link_status(dev, &link);
+       rte_eth_linkstatus_set(dev, &link);
+out_unlock:
        rte_spinlock_unlock(&hw->state_lock);
+
+       return 0;
 }
 
 static int
 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
 {
-       struct rte_eth_link link, old;
+       struct rte_eth_link link;
        uint16_t status;
        struct virtio_hw *hw = dev->data->dev_private;
+
        memset(&link, 0, sizeof(link));
-       virtio_dev_atomic_read_link_status(dev, &link);
-       old = link;
-       link.link_duplex = ETH_LINK_FULL_DUPLEX;
-       link.link_speed  = ETH_SPEED_NUM_10G;
+       link.link_duplex = hw->duplex;
+       link.link_speed  = hw->speed;
+       link.link_autoneg = ETH_LINK_AUTONEG;
 
-       if (hw->started == 0) {
+       if (!hw->started) {
                link.link_status = ETH_LINK_DOWN;
+               link.link_speed = ETH_SPEED_NUM_NONE;
        } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
                PMD_INIT_LOG(DEBUG, "Get link status from hw");
                vtpci_read_dev_config(hw,
@@ -2048,6 +2589,7 @@ virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complet
                                &status, sizeof(status));
                if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
                        link.link_status = ETH_LINK_DOWN;
+                       link.link_speed = ETH_SPEED_NUM_NONE;
                        PMD_INIT_LOG(DEBUG, "Port %d is down",
                                     dev->data->port_id);
                } else {
@@ -2058,9 +2600,8 @@ virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complet
        } else {
                link.link_status = ETH_LINK_UP;
        }
-       virtio_dev_atomic_write_link_status(dev, &link);
 
-       return (old.link_status == link.link_status) ? -1 : 0;
+       return rte_eth_linkstatus_set(dev, &link);
 }
 
 static int
@@ -2068,9 +2609,10 @@ virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
 {
        const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
        struct virtio_hw *hw = dev->data->dev_private;
+       uint64_t offloads = rxmode->offloads;
 
        if (mask & ETH_VLAN_FILTER_MASK) {
-               if (rxmode->hw_vlan_filter &&
+               if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
                                !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
 
                        PMD_DRV_LOG(NOTICE,
@@ -2081,20 +2623,18 @@ virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
        }
 
        if (mask & ETH_VLAN_STRIP_MASK)
-               hw->vlan_strip = rxmode->hw_vlan_strip;
+               hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
 
        return 0;
 }
 
-static void
+static int
 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 {
        uint64_t tso_mask, host_features;
        struct virtio_hw *hw = dev->data->dev_private;
+       dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
 
-       dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
-
-       dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
        dev_info->max_rx_queues =
                RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
        dev_info->max_tx_queues =
@@ -2102,32 +2642,35 @@ virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
        dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
        dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
        dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
-       dev_info->default_txconf = (struct rte_eth_txconf) {
-               .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
-       };
 
        host_features = VTPCI_OPS(hw)->get_features(hw);
-       dev_info->rx_offload_capa = 0;
+       dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
+       dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
        if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
                dev_info->rx_offload_capa |=
                        DEV_RX_OFFLOAD_TCP_CKSUM |
                        DEV_RX_OFFLOAD_UDP_CKSUM;
        }
+       if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
+               dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
        tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
                (1ULL << VIRTIO_NET_F_GUEST_TSO6);
        if ((host_features & tso_mask) == tso_mask)
                dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
 
-       dev_info->tx_offload_capa = 0;
-       if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
+       dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
+                                   DEV_TX_OFFLOAD_VLAN_INSERT;
+       if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
                dev_info->tx_offload_capa |=
                        DEV_TX_OFFLOAD_UDP_CKSUM |
                        DEV_TX_OFFLOAD_TCP_CKSUM;
        }
        tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
                (1ULL << VIRTIO_NET_F_HOST_TSO6);
-       if ((hw->guest_features & tso_mask) == tso_mask)
+       if ((host_features & tso_mask) == tso_mask)
                dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
+
+       return 0;
 }
 
 /*
@@ -2144,15 +2687,5 @@ __rte_unused uint8_t is_rx)
 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
-
-RTE_INIT(virtio_init_log);
-static void
-virtio_init_log(void)
-{
-       virtio_logtype_init = rte_log_register("pmd.virtio.init");
-       if (virtio_logtype_init >= 0)
-               rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
-       virtio_logtype_driver = rte_log_register("pmd.virtio.driver");
-       if (virtio_logtype_driver >= 0)
-               rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
-}
+RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
+RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);